hkuds/vimax

ViMax is a system from HKUDS that uses AI agents to generate videos from a simple idea, a novel, a written screenplay, or a personal photo. The project frames itself as four roles working together: a Director, a Screenwriter, a Producer, and a Video Generator, all combined into one tool. The stated goal is to handle the whole creative pipeline rather than just producing a few seconds of footage at a time. The README points to three problems with current AI video tools. First, most can only make very short clips. Second, characters and scenes tend to change in unpredictable ways from frame to frame, breaking consistency. Third, the tools focus only on visuals and ignore script writing, audio, and narrative structure. ViMax is presented as an attempt to cover all of those pieces in one workflow. Four main modes are described. Idea2Video turns a raw idea into a full video story by automating story design, character creation, and production. Novel2Video reads a complete novel and adapts it into episodic video content, with narrative compression and scene-by-scene visuals. Script2Video lets a user write any screenplay, from a personal story to an adventure, and produce video that follows the script directly. AutoCameo takes a photo of a person or pet and turns that subject into a recurring guest star across different scripts and scenes. The project is written in Python 3.12, uses the uv package manager, and is released under the MIT license. The README links to a Chinese version, a WeChat group, a Feishu group, and a YouTube channel, and shows several generated demo videos as proof of output quality.